Boards›Player Behavior

How the punishment system (IFS) works explained.

Ellatiel·1/23/2020, 1:44:38 AM·4 votes·6,186 views

After being banned quite a few times and reading through this board and testing some things, I believe that I have a pretty good idea how the IFS works. While the specifics may not be entirely accurate and there are some aspects that I am a bit unsure on, for the most part, this is how it works:

  • All chat logs are rated on a scale from 1-100 (or similar) with 0 being completely not toxic or rude or unsportsmanlike in any way, and 100 being the worst hate speech and toxicity you could imagine.

  • Every game you are reported, the system checks your text and assigns it a score. A game in which you are not reported means no one had any issues so you automatically get 0 points for that game (so it's best to not be reported, but false reports don't do much).

  • You receive a punishment based on reaching a constantly adjusting threshold. This is based on your toxicity score and the numbers of reports you've received on average, both of these are probably in the last X games (50, 100, running average, weighted average where recent games are weighed more, who knows). If you are frequently reported, your threshold may be as low as 40-50, while a nontoxic player might need like 70-90. If you somehow reach way above the threshold without punishment, it can give you an even stricter punishment. There is likely some kind of minor reset when reaching a punishment tier (but not too large, which is why tiers can advance so quickly if you don't change your behavior).

Some other notes:

  • ZT is separate from IFS and gives 14 day suspension/perma based on two uses of a ZT word. Intentionally feeding and throwing games are also a different system.

  • There are manual checks for people frequently reported but not punished.

  • I don't think pre-lobby chat or post-lobby chat is checked in any way, but obviously it could lead to you getting reported more which could lead to quicker punishments, or someone could submit a ticket and you could be manually punished if it was particularly egregious.

  • Now for the actual chat rating system, I'm pretty sure the IFS scoring system is little more than a word/phrase filter, and more complex things are not really taken into account (although it seems to absolutely despise all caps). Possibly, there could be words or phrases that offset toxic words and may lower your score, but not sure. Also, I do believe that the system is based upon what is reported, with words that are frequently occurring in reported games getting increased scores. This is why passive aggression and unique insults either don't get punished or get punished much slower. Now obviously, there is probably some minimum threshold a word/phrase has to be used to be included and some are probably manually added/excluded. But it changes for each region and does change over time a bit based on player reports.

  • For games shown on your reform card, I am not sure what the specifications are on this. It appears there is no minimum threshold for games to appear on your reform card (for example, one of my games literally only said "9x"), so it might just be highly reported games or something else.

What I don't know (personal experience, useful examples, and those with a history of being banned could help out):

  • How the IFS rating system for chat works specifically.

  • How the reform card games are chosen. Why is it sometimes only one mild game, why are some games chosen where there is clearly no verbal abuse.

  • If player reports have higher or lower weights depending upon reporting accuracy, or if all are weighed the same.

  • How exactly punishment tiers reduce (I went down three whole punishment tiers in about two weeks regularly flaming though, so it's pretty fast).

And Riot Tantram if you want to shoot a PM clarifying some of these aspects, that would be appreciated. Thanks [slayer-pantheon-thumbs]

4 Comments

Nightsky Pirate1/23/2020, 2:45:59 AM4 votes

It appears you're trying to tell us how the system works. Let's talk about that.

All chat logs are rated on a scale from 1-100 (or similar) with 0 being completely not toxic or rude or unsportsmanlike in any way, and 100 being the worst hate speech and toxicity you could imagine.

That's a fair enough. Might not be accurate, but it works as a model.

Every game you are reported, the system checks your text and assigns it a score. A game in which you are not reported means no one had any issues so you automatically get 0 points for that game (so it's best to not be reported, but false reports don't do much).

So Tantrum has said, who probably won't be responding to this thread...

You receive a punishment based on reaching a constantly adjusting threshold...

We've seen players go for a year without punishment and then get banned for a single slip, as provided by the sole example Riot Support gives them. We know more goes into it, but the point is there can be really long gaps and there is evidence of players being honorable for months before having a bad slipup. So, once again, probably not accurate, but makes for a good model.

ZT is separate from IFS and gives 14 day suspension/perma based on two uses of a ZT word. Intentionally feeding and throwing games are also a different system.

Not quite. They're the same system, it's like how streaming video has video and audio, two different aspects to one system.

There are manual checks for people frequently reported but not punished.

Sure.

I don't think pre-lobby chat or post-lobby chat is checked in any way, but obviously it could lead to you getting reported more which could lead to quicker punishments, or someone could submit a ticket and you could be manually punished if it was particularly egregious.

They are logged, and can be reviewed.

Now for the actual chat rating system, I'm pretty sure the IFS scoring system is little more than a word/phrase filter, and more complex things are not really taken into account (although it seems to absolutely despise all caps). Possibly, there could be words or phrases that offset toxic words and may lower your score, but not sure. Also, I do believe that the system is based upon what is reported, with words that are frequently occurring in reported games getting increased scores. This is why passive aggression and unique insults either don't get punished or get punished much slower. Now obviously, there is probably some minimum threshold a word/phrase has to be used to be included and some are probably manually added/excluded. But it changes for each region and does change over time a bit based on player reports.

It is certainly more than a filter. As a dictionary-based program (of that we are certain), at the very least it assigns values to terms and phrases. We know it considers a lot more than just that, but most of the details are left in the dark. As a machine learning program, it does adjust the values, adding terms and phrases and giving them values. There is potentially great complexity in this- there might be multiple values for every phrase, or every term might have it's own categorical values, based on what 'module' it is being tagged by, which in turn might interact with other values differently. All of which might yield a single overall toxicity value, or perhaps multiple. The point is, we don't know how complex or simple it is, but we do know that it is a dictionary-based machine learning program, which is far more than a filter.

Also; based on player's testimonials, people are punished more quickly after they have received punishments in the first place. I conclude that this means repeated use of the same terms makes the system view those same terms as solid indicators of toxicity. Unfortunately, it could mean something else- using your model, they might be lowering the bar from 0-100 to, say, 0-70. Remember, it doesn't care about how toxic a person is, just that the person is toxic. The only reason gradient values exist is because it isn't a filter. It's a mathematical judge and jury.

Then you're spot on about regional differences and reporting. There are cultural differences that strongly influence how often and for what people report for. Pure speculation, but I expect people in China report far less than Americans do, and I must leave you to imagine why (please don't say what you imagine). This is openly known about South Korea- the standards for behavior there are different from North America. This is because the machine learning is arbitrated by reports. Good job recognizing that.

For games shown on your reform card, I am not sure what the specifications are on this. It appears there is no minimum threshold for games to appear on your reform card (for example, one of my games literally only said "9x"), so it might just be highly reported games or something else.

From what I've seen, Riot Support has a slew of games to pick from, but they need only pick one to show you. However, you can request to see your data. It takes some time, but if you specifically ask for all the information, they will give it to you a month or two or three down the road.

How the IFS rating system for chat works specifically.

They probably won't ever tell us. It's probably for the best that they don't- it would be similar to showing off all the details of your national defenses to a questionable neighbor.

How the reform card games are chosen. Why is it sometimes only one mild game, why are some games chosen where there is clearly no verbal abuse.

Some forum examples are falsifications of what actually happened. Some players believe that they can get un-banned based on popular opinion and edit their logs even with screenshots, suffice to say they are entirely wrong, which is a whole discussion in itself.

The automated system is different from Riot Support's resources, it is built to be responsive to reports in a timely manner, and that system might only show one example due to a great unknown number of variables.

If player reports have higher or lower weights depending upon reporting accuracy, or if all are weighed the same.

Reports do not get graded, at all. You can report every player in every game for nonsense and the system will indiscriminately evaluate them all. However, you would be doing the community a disservice in doing so, as you would be pushing the system toward harsher judgement.

How exactly punishment tiers reduce (I went down three whole punishment tiers in about two weeks regularly flaming though, so it's pretty fast).

...What?

F1uttershy20751/23/2020, 3:06:58 AM2 votes

Issue is the system really doesn't work.

Most people who are toxic these days are usually on accounts that are Purchased from Ebay and they purposely troll. now there are a few people who are toxic on their main accounts, But the majority of people who are now a days reported are usually people on accounts with mastery's and lvls usually around 30-50

Happy Mustache1/23/2020, 1:01:25 PM2 votes

  1. Where is the list of ZT words?

  2. Explain this: https://boards.na.leagueoflegends.com/en/c/player-behavior-moderation/Nm33n3Wx-please-unlock-my-account-false-reported

Tele II1/24/2020, 6:47:46 AM1 votes

Interesting. I wish there was a bit more detail as to how you came to these conclusions. Im not really into just trusting the judgement of random strangers. You could be using very poor logic to conclude these things for all I know. Just something beyond "ive tested some things" and "been banned quite a few times" ...like, those are not indicative of any sort of deeper knowledge on their own.